Distinction between ships and icebergs in SAR images using ensemble loss trained convolutional neural networks

1Citations
Citations of this article
5Readers
Mendeley users who have this article in their library.
Get full text

Abstract

With the phenomenon of global warming, more new shipping routes will be open and utilized by more and more ships in the polar regions, particularly in the Arctic. Synthetic aperture radar (SAR) has been widely used in ship and iceberg monitoring for maritime surveillance and safety in the Arctic waters. At present, compared with the object detection of ship or iceberg, the task of ship and iceberg distinction in SAR images is still in challenge. In this work, we propose a novel loss function called ensemble loss to train convolutional neural networks (CNNs), which is a convex function and incorporates the traits of cross entropy and hinge loss. The ensemble loss trained CNNs model for the distinction between ship and iceberg is evaluated on a real-world SAR data set, which can get a higher classification accuracy to 90.15%. Experiment on another real image data set also confirm the effectiveness of the proposed ensemble loss.

Cite

CITATION STYLE

APA

Zhang, X., Liu, Y., Zheng, Y., Zhao, Z., Li, J., & Liu, Y. (2018). Distinction between ships and icebergs in SAR images using ensemble loss trained convolutional neural networks. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11320 LNAI, pp. 216–223). Springer Verlag. https://doi.org/10.1007/978-3-030-03991-2_22

Register to see more suggestions

Mendeley helps you to discover research relevant for your work.

Already have an account?

Save time finding and organizing research with Mendeley

Sign up for free